grove-account-deletion
Warn
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using the
npx wranglerCLI to interact with Cloudflare D1 databases and R2 storage. These commands are constructed by interpolating user-provided variables (e.g.,USERNAME,USER_EMAIL,TENANT_ID) directly into bash strings and SQL statements. An attacker providing a malicious identifier (e.g.,' OR 1=1; --) could potentially manipulate the SQL execution to delete unintended records. - [COMMAND_EXECUTION]: In Step 6, the skill utilizes a bash
whileloop to iterate over R2 object keys and delete them. If a tenant's media object keys contain malicious shell characters, this could lead to arbitrary command execution when thenpx wrangler r2 object deletecommand is invoked within the loop. - [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from the user and uses it to drive destructive operations.
- Ingestion points: User-provided identifiers such as username, email, and display name are collected in
SKILL.md(Step 1). - Boundary markers: There are no boundary markers or instructions provided to the agent to sanitize or validate the user-supplied identifiers before placing them in commands.
- Capability inventory: The skill utilizes
npx wrangler d1 executefor database deletions andnpx wrangler r2 object deletefor file deletions. - Sanitization: No sanitization or escaping mechanisms are implemented for the external content before it is interpolated into the shell commands.
Audit Metadata